DGX Spark 安裝 piper tts

參考 https://github.com/OHF-Voice/piper1-gpl/tree/main

參考 https://huggingface.co/csukuangfj/vits-piper-zh_CN-huayan-medium

參考 https://huggingface.co/csukuangfj/vits-piper-zh_CN-huayan-x_low

$ uv venv --python 3.13

$ source .venv/bin/activate

$ uv pip install piper-tts

$ uv pip install g2pw

$ uv pip install requests

$ uv pip install torch --index-url https://download.pytorch.org/whl/cu130

$ uv pip install unicode_rbnf

$ uv pip install sentence_stream

$ uv pip install fastapi

$ uv pip install uvicorn

$ uv pip install python-multipart

$ python3 -m piper.download_voices

$ python3 -m piper.download_voices zh_CN-huayan-x_low --download-dir models

$ ls models/

$ python3 -m piper.download_voices zh_CN-huayan-medium --download-dir models

$ python3 -m piper.download_voices zh_CN-chaowen-medium --download-dir models

$ python3 -m piper.download_voices zh_CN-xiao_ya-medium --download-dir models

$ python3 -m piper.download_voices en_US-lessac-medium --download-dir models

# for voice.synthesize, 用 curl 可以成功，但 open-webui 測試失敗

$ curl -X POST http://127.0.0.1:8100/v1/audio/speech \

     -H "Content-Type: application/json" \

     -d '{"input": "你好，這是一段測試語音。"}' \

     --output output.pcm

$ ffmpeg -f s16le -ar 16000 -ac 1 -i output.pcm \

       -codec:a libmp3lame -b:a 128k output.mp3

# voice.synthesize_wav, 用 curl 可以成功，並且 open-webui 測試成功

$ curl -X POST http://127.0.0.1:8100/v1/audio/speech \

  -H "Content-Type: application/json" \

  -d '{"input":"你好，這是 synthesize_wav 測試"}' \

  --output output.wav